EP1112621B1 - Device and method for entropy encoding of information words and device and method for decoding entropy-encoded information words - Google Patents

Abstract

A method and a device for entropy encoding and associated decoding make use of a code consisting on the one hand of a code table with reversible code words (12) and comprising on the other hand an escape region for information words to be coded which are located outside the region (14) defined by said code table. Said region can be selected in such a way that a major part of the information words is coded with symmetrical code words by the code table. On the one hand, it is thus possible to carry out, in addition to forward decoding, also backward decoding (24) and on the other hand, use of reversible code words allows for rapid recognition of errors in a code word stream transmitted over a non-ideal channel.

Description

The present invention relates to a concept for Entropy coding and a corresponding concept for decoding of entropy-encoding information words. In particular The present invention relates to the fail-safe Entropy coding and corresponding decoding of audio signals.

Modern audio coding processes or decoding processes that work according to the standard MPEG layer 3, for example able to control the data rate of audio signals for example to compress by a factor of 12 without sacrificing their quality deteriorate noticeably. To such a high To achieve data rate reduction, an audio signal is sampled, resulting in a sequence of discrete-time samples is obtained. As is known in the art, this becomes Sequence of discrete-time samples using suitable Window functions to windowed blocks of windows to obtain temporal samples. One block in time windowed samples are then by means of a filter bank, a modified discrete cosine transform (MDCT) or another suitable device in the frequency domain transformed to obtain spectral values that overall the audio signal, d. H. the timeline, given by the block of discrete-time samples is in the frequency domain. Usually will overlapping temporal blocks are generated to 50% and transformed into the frequency domain by means of an MDCT, whereby always due to the special properties of MDCT for example 1024 discrete-time samples for 1024 spectral values to lead.

It is known that the human absorption capacity Ear depends on the instantaneous spectrum of the audio signal itself. This dependency is in the so-called psychoacoustic Model, by means of which it has been possible for a long time, depending on the current spectrum masking thresholds to calculate. Masking means that a certain tone or spectral component is hidden, for example, if neighboring spectral range has a relatively high energy. This fact of masking is exploited to the spectral values available after the transformation, if possible roughly quantize. It is therefore sought, on the one hand Avoid audible interference in the decoded audio signal and on the other hand to use as few bits as possible, to encode or quantize the audio signal here. The disturbances introduced by the quantization, i. H. the quantization noise is said to be below the masking threshold lie and thus be inaudible. According to known The process is therefore divided into spectral values So-called scale factor bands performed, the frequency groups of the human ear. spectral be in a scale factor group with a scale factor multiplied by spectral values of a scale factor band to scale overall. That by the scale factor scaled scale factor bands are then quantized, whereupon quantized spectral values arise. Of course, a grouping in scale factor bands not crucial. However, it is in the standards MPEG-Layer 3 or with the standard MPEG-2 AAC (AAC = Advanced Audio coding).

A very important aspect of data reduction is in the entropy coding of the following after quantization quantized spectral values. For entropy coding usually Huffman coding is used. Under one Huffman coding is a coding with variable Length, d. H. the length of the code word for one to be encoded The value depends on its probability of occurrence. Logically you rank the most likely Character the shortest code, d. H. the shortest code word, too, so that with the Huffman coding a very good redundancy reduction can be achieved. An example of an all around well-known coding with general length is the Morse alphabet.

In the audio coding, Huffman codes are used to code the quantized spectral values used. A modern audio coder, who works according to the MPEG-2 AAC standard, used to encode the quantized spectral values various Huffman code tables that match the spectrum assigned in sections according to certain criteria. There are always 2 or 4 spectral values in one code word coded together.

A difference of the procedure according to MPEG-2 AAC compared to that Process MPEG-Layer 3 is now that different Scale factor bands, d. H. different spectral values, too any number of spectral sections or "sections" grouped become. In AAC, includes a spectral section or a "Section" but preferably at least four spectral values more than four spectral values. The entire frequency range of the Spectral values are therefore divided into neighboring sections, where a section represents a frequency band such that all sections together the entire frequency range, the by the spectral values after their transformation is covered.

A section is now just like the MPEG Layer 3 process to achieve maximum redundancy reduction a so-called Huffman table from a plurality of such Assigned to tables. In the bit stream of the AAC process, which usually has 1024 spectral values the Huffman code words for the spectral values in ascending order Frequency order. The information about that in everyone Frequency section table is used in the page information transfer.

In addition to the spectral values, the standard MPEG-2-AAC (ISO / IEC JTC1 / SC29 / WG11 IS 13818.7) the scale factors of one Huffman coding subjected to the amount of bits to be transmitted further reduce. To further increase efficiency the scale factors within a frame, i.e. H. within an associated windowed and in the frequency domain transformed blocks of samples, difference encoded. Starting from a starting value, typically the first scale factor of a frame, the absolute is given, the difference is determined. This is special efficient for compression since small changes from one scale factor to the next are very likely.

A disadvantage of the Huffman code used is the fact that it has practically no redundancy. Although this is from Bit saving or data compression reasons absolutely desirable it leads to the fact that there is no redundancy is available in order to be able to maintain a robustness against errors.

If a Huffman-encoded signal has an error Channel is transmitted, there is almost none in the decoder Possibility of still valid values after occurrence to save the error. This is based on the Huffman coded Scale factors briefly explained. As mentioned earlier the Huffman code is a variable length code. This is the very essence of the Huffman code, such that very common values associated with very short code words get while rarer values tend to receive longer or very long code words. In bitstream syntax of the audio encoder addressed are the Huffman code words in sequence for one frame in the bit stream written. The beginning of the code word for a scale factor can only be determined if the corresponding preceding one Code word correctly recognized, d. H. decoded is.

For example, this means that between 40 and 100 scale factors occur in a frame, depending on how many scale factor bands have been generated. This also means that about 40 to about 100 scale factors per frame Huffman encoded become. The code words for the individual scale factors are simply in ascending order written in the bitstream. Carries out the transmission of the bit stream via a faulty channel, such as is a radio channel, to a bit error that is the length of the code word that changed the very first scale factor is assigned, it is impossible to measure the scale factors of the decode entire frames error-free because of the decoder has no way of beginning the code word for the to determine the second scale factor. Although under certain circumstances all other scale factors except the one at the beginning Scale factor that was disturbed in the example is correct have therefore been transmitted, there is no possibility in the encoder more, the correctly transferred scale factors decode.

The specialist publication "A tool for generating bit error resilient VLC tables "by Göran Bang and by Göran Roth, Proposal for ISO / IEC JTC 1 / SC29 / WG11 dated July 1996 on a concept for encoding and decoding video information and audio information using a Variable length codes (VLC) used in conjunction with channels prone to errors. When a Bit error in the forward direction of a received bit stream is detected is decoded in the reverse direction. If a bit error occurs even when decoding in the reverse direction is detected, the decoding in the reverse direction also canceled. The one used Code is an unbalanced, fixed-length code in which a symmetrical code of variable length is mixed, such that on a certain number of bits of a code word fixed length with a bit of a symmetrical code word variable length follows. The symmetrical code words are more variable Length only serve for the robustness of errors and wear no useful information. On the receiving end, first the symmetrical variable length codewords are extracted and analyzed for transmission errors.

A disadvantage of this mixed code is the fact that Errors cannot be found in the code words fixed length occur because only the symmetrical Code words of variable length are examined. on the other hand can undisturbed fixed-length code words as error-prone can be identified if the associated code words are more variable Length are disturbed.

U.S. Patent No. 5,488,616 is concerned with a system for delivering reversible codes of variable length. For this is created from a non-reversible variable length code, the generated only provisionally, an asymmetrical reversible Code generated. The non-reversible code is more variable Length is also converted into a symmetric reversible code converted. A selector either selects the asymmetric reversible code or the symmetric reversible Code as output signal. The symmetrical reversible Code is represented by a full code tree, where all branches are either symmetrical Code words or are terminated by branch points, these branching points again by a symmetrical Codeword are completed or to further branch points to lead. The code tree contains only symmetrical code words.

EP 0 732 855 A2 discloses a system for coding and / or decoding video images using Variable length code words. The encoder includes one first encoder with a code word table for source symbols in a range of source symbols, this first Code table Codewords · with variable length, the source symbols are assigned. The source symbols, by code words of variable length of the first code table can be coded, have a relatively high frequency of occurrence. A source symbol for which no code word from the first code table is present is in a second encoder entered a code table with code words has a codeword of fixed length around the source symbol Assign length. In addition, the code word becomes firmer Length preceded and followed by an escape code, whereby the escape code from the code table of the first encoder, which has variable length code words. The code words variable length of the first encoder are reversible Code words, the code words of the second encoder one have a fixed length. This creates a single data stream generated from reversible code words of variable length and consists of escape codes, always between two Escape codes is a fixed length code word. This Data stream can be in both forward and in Reverse direction are decoded using a decoder, when it encounters an escape code, the group of bits, that follow the escape code as a code word of fixed length, since the decoder for the number of bits in the Group, i.e. informed about the length of the code words of fixed length is.

The object of the present invention is a Concept for entropy coding of information words and for decoding entropy-coded information words create that when transmitting the entropy-encoded Information words about a faulty channel enables better error detection and still one if possible provides good coding efficiency.

This task is accomplished by an entropy coding device according to claim 1, a device for decoding according to claim 10, a method for entropy coding according to claim 19 and by a method for Decoding solved according to claim 20.

The present invention is based on the finding that that only the information words are effectively transmitted in an error-proof manner can be created by reversible, e.g. B. symmetrical, Codewords are encoded. Allow only reversible code words forward and backward decoding of a sequence of code words that are a sequence of information words is clearly assigned. In contrast to the Huffman code, the has asymmetrical code words, but from a data compression point of view hub is optimal, has a symmetrical Code a higher redundancy. This redundancy can be advantageous be used for error detection. However, in the sense not too much compression gain for error protection not all according to the present invention Information words coded using symmetrical code words, just the information words contained in one certain range of information words. Information words, that are out of range not coded using the symmetrical code, but can according to a preferred embodiment of the present Invention Huffman encoded. So it becomes a Compromise between robustness on the one hand and Data compression done on the other hand.

Another important aspect for the size of the range from Information words encoded by symmetrical code words become, is the fact that for error isolation a short code, d. H. a small code table, desirable is. The size of the area implicitly defines the length of the longest code word, because with increasing number of code words in the table also the length of the valid code words increases.

According to the invention, the error limitation is carried out by that a decoder is invalid, i. H. non-reversible, Recognizes code words and deduces from this that there is a transmission error exists because such a code word by definition was not generated in the encoder. The probability that a fault leading to an invalid code word is then on highest if there are only a small number of code words is. If there is a very large number of code words, so the probability that a Fault leads to an invalid code word, since the Length of the invalid code words increases more and more.

The method according to the invention is particularly advantageous there where the information words to be encoded are essentially lie within a range, and only with a smaller one Probability information words out of range lie. The smaller the area, the less symmetrical Code words are required, and so much the better Error detection by adding artificial invalid Code words could be increased. So it tries the area of information words using symmetric Codewords are encoded in the sense of an effective one To choose error containment as small as possible, and yet so large to choose that the information words with high probability lie within this range and symmetrical be coded to provide sufficient robustness overall to accomplish.

A preferred application of the present invention is in entropy-coding scale factors of a transform-coded Audio signal, because in this application statistical seen 98% of the occurring scale factor values within of a manageable range, which is characterized by symmetrical Codewords can be encoded that do not yet have excessive length. Entropy-coded information word that is outside of this range, so an additional value is transmitted, called "Escape" becomes. The escape value is preferably Huffman encoded and separate from the symmetrically coded scale factors in the audio bit stream transfer.

The sense of escape coding according to the invention is therefore there in it, despite a relatively small RVLC table with good ones Error detection properties over a wide range of code words to be able to cover. The coding efficiency suffers from said preferred application hardly, since escape encoded Values rarely occur here.

The application of the present invention to scale factors of a transform encoded audio signal is special therefore advantageous since there are already minor disturbances in the scale factors through a non-ideal channel too cause very audible interference, as is known to be a scale factor several spectral lines multiplicatively weighted. There also the scale factors compared to the coded Spectral values only a relatively small part of the total bit amount make protection of the scale factors a redundant code does not result in significant additional effort of bits. This low additional effort is due to the error security of scale factors compared to theirs Introduce a bit amount of disruptive interference into an audio signal can, more than justified.

However, the present invention is not based on entropy coding or decoding of scale factors limited, but is advantageous wherever information words are to be coded, which are very likely to be in one Range are such that with no great loss of efficiency relatively short symmetrical code words can be managed can, values outside the range due to escape sequences can be encoded.

Fig. 1

ein schematisches Blockschaltbild für einen erfindungsgemäßen Codierer;

Fig. 2

ein schematisches Blockschaltbild für einen erfindungsgemäßen Decodierer;

Fig. 3

eine schematische Darstellung des durch den in Fig. 2 gezeigten Decodierer verarbeiteten Stroms aus Codewörtern;

Fig. 4A

bis 4C symmetrische Codes gemäß dem Stand der Technik; und

Fig. 5

einen umkehrbaren Code gemäß der vorliegenden Erfindung.

Preferred embodiments of the present invention are explained below with reference to the accompanying drawings. Show it:

Fig. 1

a schematic block diagram for an encoder according to the invention;

Fig. 2

a schematic block diagram for a decoder according to the invention;

Fig. 3

a schematic representation of the stream of code words processed by the decoder shown in FIG. 2;

Figure 4A

to 4C symmetrical codes according to the prior art; and

Fig. 5

a reversible code according to the present invention.

Before going into the figure in detail, however, some general aspects of entropy coding are presented. In particular, the special features of the Coding of differentially coded values received on advantageous way with the coding by means of symmetrical Code words and escape values can be combined.

The code according to the invention represents an entropy code, often similar to the widely used Huffman code short code words and rarely occurring values Assigns long code words to values. The code according to the invention however differs from the Huffman code in the following ways:

First, the coding used, in contrast to the Huffman coding a decoding from both sides (forward and backwards). This is also called in technology "Reversible Variable Length Coding" (RVLC). The Backward decoding is possible if reversible, e.g. B. symmetric code words, d. H. when a Code book or a code table with reversible code words is used.

Furthermore, the use of an RVL code allows the effect that the code table, i.e. H. the amount of available Code words, does not form a "complete tree". It So there are code symbol sequences that are not a valid code word result, d. H. the codewords result in not being symmetrical are. The gaps that arise in the tree can be exploited to detect transmission errors as they indicate that a received code word is never indicated by a transmitter has been generated.

Since the Huffman coding is optimal in terms of information theory it doesn't make sense to use a different code if the only goal is to increase data compression maximize. However, when it comes to an increase in insensitivity to errors with as little loss of compression efficiency as possible the Huffman code is not advantageous. By coding with variable length code words an error from the disturbed code word to all subsequent ones Propagate code words. So by a bit error falsifies the length of a code word, which leads to the fact that all subsequent code words in a code word sequence are not can be decoded more because the decoder has no knowledge of which has where the corresponding code words begin.

Instead of the variable length code, e.g. B. the Huffman code, now uses a reversible code of variable length, an error that changes the length of a code word can usually be diagnosed very quickly due to the presence of invalid code words. If an invalid code word becomes known, the decoding is aborted. This is not possible with Huffman coding because all code words are valid and therefore no invalid code words exist. In addition, the RVLC allows decoding from behind, which makes it easier to isolate the error. The following example shows this. A code table according to the invention could, for example, be as follows: information word code word 3 110011 2 11011 1 111 0 0 -1 101 -2 1001 -3 10001

The area for information words through this code table be coded using only symmetrical code words can range from -2 to +2. Values outside of this There are limits, d. H. which are greater than 2 in amount, in addition to the symmetrical code word for -3 or Escape +3.

10000

11000

110010

11010

Code words that cannot appear in the code table shown are the following:

10000

11000

110010

11010

With regard to a more detailed representation of an inventive Codes are shown in Figure 5 and the corresponding discussion referenced later.

In the following, let us consider a sequence of numbers 1, -3, 1, 2, -1 as a fictitious example, which is to be transmitted via a faulty channel: Sequence of information words 1, -3, 1, 2, -1 Sequence of code words 111 10001 111 11011 101

If the case is considered that the twelfth bit is disturbed by an error generated by the channel, the following bit sequence results:
disturbed sequence of code words: 111, 10001, 111, 0 1011, 101

Decoding five values from the front gives 111, 10001, 111, 0 , 101, ie 1, -3, 1, 0, -1.

However, the decoding from behind gives the following sequence:
101, 11010,
ie only -1 and an invalid code word. From this simple example it can be seen that the invalid code word 0 1011 is recognized very quickly by the decoding from the rear, ie by the reverse decoding. Furthermore, the error can be very quickly localized and identified by aborting the decoding process after the invalid code word. Reverse decoding therefore reports an error in the range from the last bit to the eighth bit from behind.

The two decoded series of numbers are as follows. A bold print in the table below means that the values can be incorrect: forward 1 -3 1 0 -1 backward x x x x -1

The extent to which an error can be narrowed down depends on the type of error from and on the implemented error concealment technique. Known error concealment techniques consist of simply replacing a faulty one Value by its neighboring intact value. on the other hand can, if both intact values that border on an error, weighted averages from the left and right margin used to the incorrect value to replace artificially, d. H. to disguise. Still others Error concealment techniques use interpolation using two adjacent values between which there is a mistake. Likewise, one-sided prediction be taken from the front or from behind to make a faulty one Value through a "possibly relatively intact" Value to replace.

1 shows an entropy encoder according to a preferred one Embodiment of the invention. In a decision maker 10 information words to be coded are fed in. In the The tables above consisted of information words of simplicity for the sake of integer numbers only. With an audio encoder are information words that are entropy-encoded should be such. B. scale factors, for example as There are eight bit values. The expression "information words" should therefore encompass every type of presentation, through which one coding information can be displayed.

It is determined in the decision maker 10 whether the information word in a range of information words or outside the Range of information words. The area will determined by the code table implemented in the encoder. If it is determined in the decision maker that a code to be coded Information word is within the range the same to a device 12 for assigning a symmetrical one Code words from a group of symmetrical code words, d. H. from the code table, transmitted such that a symmetrical code word is assigned to the information word becomes. On the other hand, the decision maker 10 decides that the Information word outside of the one determined by the code table Area, this information word is from Decision maker 10 for a device 14 for generating a added value, such that the facility 14 in a preferred embodiment of the present Invention determined the escape value. The facility 14 basically comprises two outputs, i. H. an exit to Writing the escape value into an escape area of the bit stream, and, on the other hand, an output connected to a bitstream formatter 16 which is connected to a current of Code words or a sequence of code words generated that the sequence of Information words is assigned.

For a more detailed explanation of how the facility works to generate an additional value or escape value, that are identified in FIG. 1 by reference numeral 14 is discussed in Fig. 3. Fig. 3 shows one continuous stream 30 of "potentially" symmetrical code words, the term "potential" is intended to indicate that the stream 30 already over a non-ideal channel, e.g. B. a radio link has been transmitted, causing bit interference may have occurred. The stream consists of individual symmetric code words 30a, 30b, 30c, etc., all within of through the code table, the symmetrical code words includes, specified range. The electricity out however, potentially symmetric code words also includes symmetrical code words 31, 32 for information words stand at the edge of the area. The code words 30a-30c become generated by the allocation device 12 and in the Bitstream formatter 16 fed. The code words for Words of information at the edge of the area will appear at a preferred embodiment of the present invention generated by the device 14 and from there in the Bitstream formatter 16 is fed, the one shown in FIG Stream 30 forms. Codewords 30a-30c and 31 and 32 represent information words from -7 to +7, i. H. Information words to which symmetrical code words are assigned are. Has the information word to be coded a value of for example +12, the sum is the symmetric Code word 31 and the escape value the value +12.

A decoder that encounters code word 31 immediately recognizes that this is a code word at the edge of the range acts, which is why the decoder for decoding the information word using link A in the escape area "goes" to find that an escape value of im present example 5 is present. The facility 14 leads to generation according to a preferred embodiment the present invention thus has two functions by. First, it provides the code word for the edge of the Range in the stream 30 from symmetric code words. To the others it is the difference between the one to be encoded Information word and the code word at the edge of the area and creates an escape value that represents the difference. Of course, the escape value can be changed using the coding methods according to the invention are entropy-coded. However, it is preferred for data compression reasons that To encode the escape value using a Huffman code. From Fig. 3 it can also be seen that the escape value is not in the Stream is written from symmetrical code words, but somewhere else in the bitstream.

If the value -12 is now to be coded, the decision maker becomes 10 find that this value is outside the range defined by the Code table with symmetrical code words defined area lies. The device 14 for generating an additional On the one hand, value becomes the code word for the value -7 output to bitstream formatter 16 and the other the difference, d. H. 5, write in the escape area. The Value -12 is then the result of the combination of the value -7, in FIG. 3, for example, the code word 32 and the escape value 34 via the link represented by arrow B.

A value of +7 would be in the preferred embodiment of the present invention as a code word for +7, i.e. H. as the code word 31, and 0 in the escape area, i.e. H. as the Escape value 33, can be encoded.

In deviation from the described embodiment, it is not necessarily that the device 14 for generating a additional value is the difference between the information word to be encoded and the information word at the edge of the area forms and on the one hand a symmetrical code word in the Stream 30 of symmetric code words writes and on the other hand writes the difference in the escape area. It is alternatively also possible that the entire information word in the escape area is written, and in the stream of symmetric Code words just a placeholder, a specific one Bit combination or the like either by the Device 14 or by the bit stream formatter 16 is inserted to a downstream decoder signal that it is at this point in the bit stream in the Escape area must go. The procedure shown has but the advantage that at least the part of the information word, which is in the range defined by symmetric Code words is covered by means of a symmetrical code word is encoded, leaving only the difference, for example is encoded using a Huffman code, less is strongly secured or error-proof. The alternative procedure would have the advantage, however, that no addition or Difference formation must be carried out, and that for a Information word out of range of less redundant Code is used. The disadvantage, however, is that the information word out of range is not can be decoded backwards. However, it can either processes known in the prior art or the process according to the invention Coding method used to protect the escape area to create safe conditions here too.

Fig. 2 shows a preferred embodiment of a decoder according to the present invention. A sequence Codewords or a stream of "potentially symmetrical Code words "30 are fed into a memory 20 which an analyzer 21 can access to one in memory Analyze 20 saved sequences. The analyzer 21 comprises a device for detecting a symmetrical code word from the sequence 30 of code words and on the other hand, means for detecting a predetermined one Codes in sequence 30 of code words. Detected the analyzer 21 an intact symmetrical code word, so it transmits this to a device 22 for assignment a certain information word based on the code word on a previously known code table, that of the encoder (Fig. 1) used code table must correspond. Detects the Analyzer 21, however, has a predetermined code that is present Example the code word for an edge of the area is, he gives this code word to a device 23 for Finding an additional information word outside of Range. The device 23 is in the preferred embodiment when a code word 31 occurs in the stream 30 access the escape area and there the corresponding Get the escape value and go to the information word that corresponds to code word 31, add or subtract.

The assignment of a predetermined code, in the exemplary embodiment the code word for an information word on the edge, to a Code word in the escape area can be in different ways and wise things happen. The easiest is to use one consecutive pointer, being both the escape area as well as the stream 30 are synchronized. This synchronization is in the preferred embodiment of the present Invention in which scale factors are encoded entropy are given by the fact that always a block or frame of a Audio signal is processed. For each block the Escape area and stream 30 of symmetric code words initialized so that the use of a continuous Pointer provides correct results.

The encoder according to the invention further comprises a device 24 to reverse the order of the decoder by the analyzer 21 is activated. The analyzer finds 21 an unsymmetrical code word in the stream 30, so it becomes the Activate device 24 for reversing the order, since an asymmetrical code word in the stream 30 from potentially symmetric code words cannot occur. The facility 22 to assign and the device 23 to determine then work in reverse order from the other end the sequence of code words to be decoded with narrow down the error in such a way that only possible few values can be replaced by an error concealment have to.

In reality it can happen that an entropy decoder a disturbed code word does not immediately recognize because of the disturbance did not result in an invalid code word. He decodes So beyond the error until it is caused by a subsequent error encounters an invalid code word and the coding then breaks off. The backward decoder then decodes from other end also possibly over the disturbed Code word and breaks at some point in the case of an invalid one Code word. This created an overlap area in which both the entropy forward and the entropy backward decoders Have delivered initial values. So that is the error is narrowed to the overlap area, and it it can be determined that the decoded values are outside of the overlap area are correct.

If differential coding has been carried out in the encoder is, the corresponding decoder further comprises one Differential decoder 25, which is the differential coding generated in the encoder undo again. The differential decoder 25 is according to the invention by the device for reversing order 24 also activated to a Reverse differential decoding to perform differential encoding Information words used by the institution 25 transmitted by the facilities 22 and 23, also in Backward direction again completely decoded information words manufacture. It should be noted that the Reverse differential decoder and the forward differential decoder may be separate facilities or the same are implemented by a single facility, with an addition in forward differential decoding is executed while in reverse differential decoding a subtraction is performed.

The combination of differential coding with the coding method According to the present invention, it is particularly advantageous only suitable by a differential coding at selected starting value of the difference formation the absolute Information words, e.g. B. by a symmetrical to zero Area to be "moved".

However, this means a differential decoding 25 in the reverse direction from the other end of an information word sequence is possible must at the end of the sequence of information words an additional value is added in the encoder, such that the differential decoder knows from where a differential decoding should be started from behind. If a fixed start value is used for differential coding has become an additional value additional difference value at the end of the sequence of difference codes Informational words indicating the difference from that last information word for the specified or predetermined Indicates the starting value. The difference becomes self-evident also entropy-encoded and is preferably by means of a symmetrical information word entropy-coded, such that this value is well protected for backward decoding to allow. If as a starting value for the differential coding the first information word in the encoder Sequence of information words is taken, it offers the absolute value as an additional value at the end of the sequence of the last information word. This then last value will almost certainly not be in the range of information words lying with symmetrical code words be encoded.

As previously mentioned, there is a preferred application of the present invention in the encoding of Scale factors that were previously differential first and then Huffman encoded. In the prior art, one Huffman table with 121 code words used to find values in Range from -60 to +60. Because the number the scale factors to be coded compared to the number the spectral values are very small, a typical value lies at 40, is a relatively "fast" detection of errors absolutely necessary, such that the decoder after the Decoding fewer values aborts so an error is relative can be narrowed down well. Therefore a "small" Code table used, which means that the number of symmetric RVLC code words should be small.

The smaller the code table or code book, the more earlier detection of errors is more likely. Therefore, the area for information words runs be encoded symmetrically, from -7 to +7. For values that Range from 7 to 60, the escape value is transferred. Preferably the same Huffman is encoded. The "Escape" table consists of 54 entries for values between 0 and 53. Every time the recipient receives a -7 or decodes a +7, it must have the associated escape value decode and add or subtract.

Statistically speaking, the interval -7 to +7 covers 98% of the occurring scale factor values, such that escape values not particularly common. Should be more common escape values occur, or will fail safety one more time More attention is given to various known ones Process and also the inventive method used to make the escape values more error-proof.

To explain the reversible code according to the invention with variable length is first discussed in Fig. 4A, in the a known symmetric code is shown, for example in the aforementioned specialist publication by Göran Bang and Göran Roth is revealed. This code is defined by a code tree that has a root 40 and a Has branch point 42. Are at the root of it is a binary code, two branches 43, 44 are attached, where branch 43 is root 40 with an end point connects that defines the valid code word "1". The branch 44 connects the root 40 to the branch point 42, from the two branches 45, 46 lead away. The branch 46 is with one Connected endpoint which is the second valid code word this Codes "00" defined while branch 45 is an invalid one Codeword of this code, i. H. "01" defined. The code word "01" is invalid because it is asymmetrical. In terms of the further notation it should be noted that in 4A and 4C and in Fig. 5 invalid code words are framed are. The code shown in Fig. 4A thus comprises only two valid code words, i. H. "1" and "00", and only one invalid code word, i. H. "01", the same long as the second valid code word is "00".

A slightly longer code is shown in Fig. 4B. In contrast 4A, the code in FIG. 4B contains another valid one Code word "010" and also an invalid code word, the now also like the additional valid code word 3 bit is long and has the value "011". In contrast to Fig. 4A branch 45 is not connected to an end point, but rather with another branch point 47, of which two branches going out, with the first outgoing branch for additional valid code word "010" is sufficient, while the other branch to only invalid code word "011" is sufficient.

4C is the logical continuation of FIGS. 4A and 4B since the branch point 47 now has a further branch point 48 is connected, of which in turn two branches going out, one branch an additional symmetrical Codeword "0110" defined while the other's end point Branch defines the only invalid code word "0111" that has the same length (4 bits) as the longest code word of the Code tree, d. H. the code table.

From Figure 4C it can also be seen that there are no valid code words of equal length exist. This also applies the codes in Figs. 4A and 4B.

The codes shown in Figs. 4A to 4C are used in the Professional publication by Göran Bang and Göran Roth only as a security grid and not as a code for coding Information used as such code as it is without further by logical continuations of those shown in the figures Codes can be seen with a sufficiently high number of Code words becomes very long. In addition, the robustness of errors correspondingly longer codes very small, because only ever there is only one invalid code word and also this invalid code word as long as the longest valid one Is codeword. A coding of information with the known Codes are therefore not recommended because the code words are very become long when there is a reasonable range of information values should be encoded, and since only an invalid one Codeword exists, which is also very long. A decoder will therefore not immediately discover a mistake and commit a large number of consequential errors before going on one invalid code word occurs and decoding stops. The Errors are therefore poorly localized.

The code table according to the invention shown in FIG. 5 overcomes these disadvantages in that at least in the code tree there is a branch point, of which two branches go out, both with a branch point instead of are connected to an end point. In the one shown in Fig. 5 preferred embodiment for the reversible code Variable length, these are branch points 50a, 50b, 50c and 50d. The code tree in Fig. 5 also includes a root 52, from which two branches 53 and 54 start, the branch 53 is connected to an end point through which the first and shortest code word 0 is defined, the preferred one Embodiment of information value "0" is assigned. The code word 0 has the shortest length and is therefore in the sense an entropy coding assigned the information value, the most common. In the preferred embodiment of the present invention in which the information values are differentially coded, it has been found that especially in the differential coding of scale factors the value 0 is most likely to occur.

The other branch, which starts from the root 52, i. H. the According to the invention, branch 54 does not end in a code word of length 2 bits but leads to branch point 50a, which in turn has two branches 55, 56 with further connection points 57, 58 is connected. The connection points 57 and 58 are in turn via corresponding branches 59, 60 with Endpoints that contain the valid code words 101 and 111 define. In contrast to the state of the art, here too see that by foregoing a code word with a length of 2 bit 2 code words can be obtained which are the same are long, d. H. 3 bits in Fig. 5. These are the code words 101 and 111. They are in the code table shown in FIG. 5 assigned to the information values "-1" and "1". Under the In terms of entropy coding, it is convenient for two Informational values that are very likely to be equally common occur to provide code words of the same length.

From Fig. 5 it can be seen that from the connection point 50c two branches 62, 63 go out, the branch 63 with one Connection point 64 is connected to one with branches Valid code word 110011 and with an invalid code word 110010 is connected. For example, a decoder now hits to the invalid code word 110010, it becomes the coding process cancel because such a code word is irreversible and has never been generated in an encoder.

The formation and assignment of further code words Information values can be seen from FIG. 5. However, it is pointed out another peculiarity. A preferred one Use of the code according to the invention shown in FIG. 5 consists in using the reversible code with escape values. As it has already been shown is an information value that is out of range from "-7" to "+7", by the code word for the corresponding one Information value at the edge of the area and the further difference in an escape table coded. Therefore, there is a higher possibility that in the encoded bitstream the value at the edge of the range, i.e. H. "-7" or "7" must be coded. Furthermore, the probability same size that "-7" or "+7" occurs. According to a preferred embodiment of the present invention the information values "-7" and "+7" Code words of equal length, i. H. 1000001 or 1100011, coded, these code words being shorter than the longest occurring code words, which in this case are the Code words for "-6" and "+6" are used to create a code table create the best possible from an entropy point of view is.

From Fig. 5 it can also be seen that eight invalid code words 66a - 66h exist, while a reversible code according to the state of the art only ever one bad code word has or can have. A large number of invalid ones Code words and particularly relatively short invalid code words, such as. the code words 66e, 66f provide a high level of error resilience, such that a decoder as quickly as possible the decoding aborts after a disturbed information value, such that an error is narrowed as narrowly as possible can be.

Finally, it should be summarized that the one shown in FIG. 5 reversible variable length code particularly good for a robust robust entropy coding of information values is suitable because, on the one hand, a relatively large number short invalid code words exist, and there is another for the price of a shorter code word (in the example, that Code word "11") two code words (in the example 101 and 111) can be obtained, which is longer but equally likely are. Although the waiver of a valid short code word in Entropy coding is actually to be avoided provides the same according to the invention in applications in which error-resistant entropy coding can be performed should, and for which two additional information values with relatively high probability and especially with approximately equal likelihood of occurring, a good one Solution. Regarding the code table representation in tree form this waiver is accomplished by the fact that special Branch points exist from which two branches start, but both with one endpoint instead of another Branch points are connected.

Claims (20)

1. A device for entropy encoding of information words, comprising:

   means (12) for assigning a reversible code word from a group of reversible code words to an information word which is located within a region of information words, the group of reversible code words being designed such that, for each information word within the region, there is provided one specific reversible code word each;

   means (14) for producing a predetermined additional code word (31, 32) for an information word which is located outside the region of information words, for producing an additional value (33, 34) for the information word located outside the region of information words, and for assigning a non-reversible variable length code word from a group of non-reversible variable code words to said additional value (33, 34), with an escape region being produced that is separated from the data stream of reversible code words and consists of non-reversible variable length code words, there being always assigned one code word of the escape region to one of the predetermined code words (31, 32) of the data stream of reversible code words each (A, B); and

   data stream formatting means (16) connected to the means (12) for assigning and to the means (14) for producing and being arranged to produce a data stream of reversible code words comprising only reversible code words for information words within said region and comprising predetermined reversible code words for information words outside said region.
2. A device according to claim 1, wherein the reversible code words are symmetrical code words.
3. A device according to claim 1 or 2,
   wherein the means (14) for producing is arranged such that it produces an additional value also for an information word located on a boundary of said region of information words and such that it produces the predetermined code word for the information word located on the boundary of said region of information words.
4. A device according to any of the preceding claims, which is designed such that information words occurring with a probability of occurrence that is above a predetermined limit probability, are located in said region.
5. A device according to any of the preceding claims, wherein the means (12) for assigning is arranged such that the size of the region is selected so that the length of the longest code word of the group of reversible code words is smaller than or equal to a predetermined length defined by error recognition aspects.
6. A device according to any of the preceding claims, wherein the means (14) for producing is arranged so as to form as additional value the difference between the information word and the information word on a closest boundary of said region.
7. A device according to any of the preceding claims,
   wherein the information words each correspond to numerical values from a first to a second boundary and numerical values outside the boundaries, said first and second boundaries defining said region of information words,
   wherein the means (12) for assigning is arranged to assign to an information word located on a boundary or outside said region, the predetermined reversible code word, the predetermined reversible code word corresponding to the reversible code word provided according to the code table for the information word corresponding to the boundary of the region which, on the basis of its numerical value, is closest to the numerical value of the information word; and
   wherein the means (14) for producing produces the difference between the information word to be coded and the information word on the boundary of the region as an additional value, the sign of the difference being determined by the sign of the information word on the boundary of the region, such that the information word to be coded is represented by the predetermined reversible code word for the information word on the boundary of the region and by a sign-free difference as additional value (33, 34).
8. A device according to any of the preceding claims, further comprising:

   means for differential encoding of information which, on the basis of a starting value, produces differential-encoded information words, with successive differential-encoded information words representing a differential value sequence.
9. A device according to claim 8, wherein the means for differential encoding further comprises:

   means for adding an additional element to the end of the differential value sequence, the additional element being defined such that backward decoding of the differential value sequence can be carried out.
10. A device for decoding of information words that are entropy-encoded using reversible code words, said information words being present in the form of a data stream and an escape region, the data stream comprising only reversible code words for information words with a region of information words and the escape region comprising non-reversible code words for additional values representing information words outside the region of information words, there being always assigned one code word of the escape region to one predetermined code word (31, 32) of the data stream each (A, B), said device comprising:

    means (21) for detecting a reversible code word (30a, 30b, 30c) from the data stream;

    means (22) for assigning a specific information word to the code word detected from the data stream, on the basis of a code table;

    means (21) for detecting a predetermined reversible code word (31, 32) in the data stream;

    means (23) for ascertaining a non-reversible variable length code word in the escape region, which is assigned to the predetermined reversible code word detected from the data stream, for assigning an additional value to the non-reversible variable length code word on the basis of an escape table for the non-reversible variable length code words, and for ascertaining an information word outside of the region of information words on the basis of the additional value.
11. A device according to claim 10, wherein the non-reversible variable length code words are Huffman code words.
12. A device according to any of claims 10 or 11, wherein the means (21) for detecting a reversible code word is arranged such that a non-reversible code word in the first data stream (30) can be ascertained.
13. A device according to claim 12, further comprising:

    means (24) for reversing a sequence in which said means (21) for detecting processes the code word sequence (30), said means (24) for reversing being responsive to a non-reversible code word being ascertained.
14. A device according to any of claims 10 to 13, wherein the information words are differential-encoded and which further comprises:

    means (25) for differential-decoding the differential-encoded information words.
15. A device according to claim 13 or 14, wherein the sequence of information words, starting from a starting value, is differential-encoded, the sequence furthermore having at the other end thereof an additional value which is selected such that backward differential decoding from the other end can be carried out, said means (25) for differential decoding being arranged to carry out differential decoding from the other end responsive to said means for reversing the sequence.
16. A device according to any of claims 10 to 15, wherein the predetermined reversible code word (31, 32) is the code word assigned according to the code table to an information word on a boundary of the region of information words.
17. A device according to claim 16, wherein the means (23) for ascertaining is arranged to form the sum of the information word on a boundary of the region and of the additional value, so as to obtain an information word located outside said region.
18. A device according to any of the preceding claims, wherein the information words are scale factors of a transformation-encoded audio signal.
19. A method for entropy encoding of information words, comprising:

    assigning (12) a reversible code word from a group of reversible code words to an information word which is located within a region of information words, the group of reversible code words being designed such that, for each information word within the region, there is provided one specific reversible code word each;

    producing (14) a predetermined reversible code word (31, 32) for an information word located outside said region of information words,

    producing (14) an additional value (33, 34) for the information word which is located outside said region of information words;

    producing a data stream of reversible code words comprising only reversible code words for information words within said region and comprising predetermined reversible code words for information words outside said region; and

    coding the additional value (33, 34) by assigning a non-reversible variable length code word from a group of non-reversible variable length code words to the additional value (33, 34), with an escape region being produced which is separated from the data stream of reversible code words and consists of non-reversible variable length code words, there being always assigned one code word of the escape region to one of the predetermined code words (31, 32) of the data stream each (A, B).
20. A method for decoding of information words that are entropy-encoded using reversible code words, said information words being present in the form of a data stream and an escape region, the data stream comprising only reversible code words for information words within a region of information words and the escape region comprising non-reversible code words for additional values representing information words outside the region of information words, there being always assigned one code word of the escape region to one predetermined code word (31, 329 of the data stream each (A, B), said method comprising the following steps:

    detecting (21) a reversible code word (30a, 30b, 30c) from the data stream;

    assigning (22) a specific information word to the code word detected from the data stream, on the basis of a code table;

    detecting (21) a predetermined reversible code word (31, 32) in the data stream;

    ascertaining a non-reversible variable length code word in the escape region, which is assigned to the predetermined reversible code word detected from the data stream;

    assigning an additional value to the non-reversible variable length code word on the basis of an escape table for the non-reversible variable length code words; and

    ascertaining (23) an information word outside of the region of information words on the basis of the additional value.

Images